“Hopelessly idealistic.” “Out of touch with reality.” “Embarrassingly inept.” This is how the OpenAI board has been described in online forums and social media this past week. Indeed, the debacle around the firing and then subsequent rehiring of CEO Sam Altman, with two interim CEOs in between, betrays an air of desperate amateurishness. Â
The latest update to this still-unfolding saga is that three of the four OpenAI board members who voted to remove Altman and demote its president and board chairman Greg Brockman have themselves been removed from the board. The three are OpenAI’s chief scientist Ilya Sutskever, robotics engineer and entrepreneur Tasha McCauley, and Georgetown University Center for Security and Emerging Technology head Helen Toner. Â
All three are known to be concerned about the potential negative effects of AI. As recently as last month, Toner called on the US government to “take action to protect citizens from AI’s harms and risks.” The fourth previous OpenAI board member, Quora CEO Adam D’Angelo, will remain on the board. Â
Just because the four acted clumsily does not mean they were wrong.Â
The main role of a corporate board is to represent the interests of shareholders. In the case of OpenAI, it is slightly different because OpenAI is nominally a non-profit foundation. The board is tasked with a fiduciary duty to uphold the organization’s core mission, set out in 2018, to build artificial general intelligence (AGI) that is safe and benefits all of humanity. AGI, in turn, is defined as highly autonomous systems that outperform humans at most economically valuable work.Â
But by 2023, the board was overseeing a vastly different organization. Driven by Altman, OpenAI had expanded massively, launching a series of products and platforms. Many of these, like ChatGPT and Dall-E, had become phenomenally successful. It had also spawned a capped-profit subsidiary (at 100X investment), 49% of which had been sold to Microsoft for $10bn. Â
“When it comes to AI, we may not have the luxury of waiting a few years to see what the private sector can do on its own before the government establishes ground rules.”
We could rightly criticize the OpenAI board for how they handled the situation, but we shouldn’t fault them for their reasons for acting. If the majority of members truly felt that Altman was steering the organization too far from its mission, then they were right to intervene. Court judges do not rule based on personal feelings; they rule based on the law. If you don’t like their decisions, then challenge the laws, not those applying them. The same logic applies to OpenAI and its board. Â
It seems likely that the board had become increasingly uneasy with the direction that OpenAI was taking as it steadily diverged from its stated mission. Last Friday, perhaps due to evidence of a new AGI-like breakthrough, these tensions came to a head. Â
It is not yet clear who the winners are in this saga. You could point to Microsoft for gaining influence among OpenAI executives and engineers. Or Google, for seeing its main AI competitor wobble uncontrollably before its eyes. Or Altman and Brockman, for cementing their statuses as AI legends. Â
While the winners can be debated, there is no question about the loser – AI safety.Â
By bungling the governance so badly, the OpenAI board scored a howler of an own goal. While their objective was apparently to slow AI development down to ensure a safe deployment, the result is likely to be the opposite. Â
AItman’s camp of AI accelerators and optimists, supported by Microsoft, has emerged squarely on top. In a post-OpenAI Silicon Valley, anyone who has serious concerns about AI safety will now have a hard time being taken seriously. The new OpenAI board members, parachuted in to steady the ship, are corporate and government elites. The adults have taken over, and they are likely to support the expansionist and competitive goals championed by Altman. Â
OpenAI’s convoluted structure – a non-profit focused on the benefit of humanity overseeing a for-profit intent on leading in an industry Bloomberg estimates will grow to $1.3 trillion over the next decade – was an earnest but futile attempt at corporate self-regulation. In many ways, it was the most audacious, grandiose, and perhaps most foolish version of Silicon Valley’s go-to response to any kind of concern about the societal impact of technology and innovation.Â
A generation ago, tech entrepreneurs convinced the Clinton Administration that businesses could self-regulate the internet. An eBay score is surely more effective than liability law. If consumers don’t want to share their personal data, they can just opt out. It isn’t music piracy if you are merely sharing your favorite tunes with your peers – even when you have 100,000 “peers”. Why should it be up to tech companies to protect minors from harmful online content when that’s clearly what parents are for? And let’s not get hung up on the law if Meta’s own Oversight Board can “answer some of the most difficult questions around freedom of expression online: what to take down, what to leave up, and why.
The history of the internet shows that self-regulation alone is usually not enough.
As business school professors, we’re sympathetic to the idea that private sector governance is often preferable to heavy-handed government intervention. That’s particularly true for fast-evolving new technologies. But the history of the internet shows that self-regulation alone is usually not enough. Business doesn’t have a great track record of policing itself (remember FTX?), and the enormous profits that frequently await tech first-movers mean that other considerations often fall by the wayside regardless of initial good intentions. It may take public officials a few years to elbow their way into new industries but if the stakes are sufficiently high, sooner or later the government will assert itself.Â
The problem is, when it comes to AI, we may not have the luxury of waiting a few years to see what the private sector can do on its own before the government establishes ground rules. The stakes are too high. The technology is too powerful. The fortunes to be made are too vast. Self-regulation will not suffice. We need to rein in the potential harm that AI can cause, and that means we need government regulation now.Â
Unfortunately, the regulatory route today doesn’t look promising. Â
Despite a strong statement noting the power and danger of AI from an international group including the EU, the US, and China earlier this fall, no concrete actions were taken, or even recommended. The UK has stated publicly that it will not regulate AI in the short term. The EU’s AI bill has also hit a roadblock, as France, Germany, and Italy are no longer happy with its attempt to slow down AI development. The US rushed out an Executive Order on AI that got a lot of pushback as being soft on big tech. Â
Looming large over Western governments is that regulation could hamper the West’s ability to keep up with China in what already feels like an AI arms race (notably, back in June, now-ex-OpenAI board member Toner penned a Foreign Affairs piece arguing that China was actually trailing and that “regulating AI will not set America back in the technology race”).
“While the winners of the OpenAI saga can be debated, there is no question about the loser – AI safety.”
Ironically, many Western countries are all too eager to restrict the transfer of sophisticated hardware across borders (think Nvidia chips or ASML lithography machines) but seem unwilling to afford the same attention to powerful software and algorithms these systems enable. Â
AI regulation is urgently required, and there are at least seven distinct areas that should be included in any legislation. The objective should be to ensure the protection of individual rights, and, ultimately, the continuation of humanity.Â
AI should not create harm to people or society.
Owners and developers of AI need to take responsibility for any adverse impacts.
AI should take active steps to minimize biases and treat all people equally.
When an output is created by AI, it should be clearly labeled as such.
AI systems should be designed to enable researchers to study how these systems operate, make decisions, and reach conclusions.
AI systems should include safeguards so that the underlying data and technology cannot be stolen and are safe from attacks.
AI should respect the privacy rights of individuals and organizations and avoid the use of personally identifiable information.
The situation with OpenAI appears to be a clear victory for AI accelerators over AI ethicists. We are boldly moving into an exciting future of AI-fueled productivity gains. But we need to ask, in the absence of formal regulation and oversight, who is keeping an eye on the dark side?
Generative AI offers organizations the opportunity to create economic value out of their data but a significant volume of that data – so-called "dark data" – goes unused, constituting a massive environmental...
Hiring a Chief AI Officer is all the rage, but not all organizations need such a role. Whether yours does will depend largely on where you are at in terms of working...
When the British Library fell victim to a major ransomware attack it opted for a policy of full transparency. Here we outline what government organizations, NGOs, and businesses can learn from the...
Not all risks posed by GenAI are the same. Here’s how to identify the four distinct types and manage them accordingly ...
Explore first person business intelligence from top minds curated for a global executive audience
